Algorithms Inspired from Active Electro-location Behavior of Weak Electric Fish, Developed for Autonomous Eel-like Swimming Robot

In this article several algorithms for detection, localization and classification of objects from measurements of electric signals are proposed. The algorithms are biologically inspired from electric fish electrolocation behavior and required minimal computational power in the way that could be easily implemented in an autonomous robot. On the lower level the electrical signal is digitalized, filtered with FFT and correlated on Reichardt type correlators [1][2]. We improve the original Reichardt algorithm by adding an adaptive mechanism, which will account for the snake like movements of the fish robot. The proposed Adaptive Reichardt Correlator (ARCO) is generic and could be used with different oscillatory locomotion patterns. On higher level we used set of ARCOs for implementation of 3 different activities – object detection, localization and shape recognition. Object detection is performed directly from the set of ARCOs, while Reinforce Learning is used for training a discriminator for localization and shape recognition of simple objects like sphere and cube. All activities are organized in Subsumption Architecture [3].